cfierro/othello-llama3-multi-format

Othello (synthetic) (Llama-3 multi-format, input_output) Continued-pre-training version of cfierro/othello-synthetic, formatted for Axolotl's template-free input_output loss masking with Llama-3 BOS/EOS. Each game is written in one of several randomly-chosen move-sequence formats so the model is robust to how games are notated. Formats Every row shares the same masked prefix <|begin_of_text|>Game of bjk\n\n; only the formatting of the move sequence (the trained… See the full description on the dataset page: https://huggingface.co/datasets/cfierro/othello-llama3-multi-format.